21(27)
Looking more closely at the data, we noticed that some households
adjusted their household composition, moving into the overcrowded
category by “adding” children to the household, i.e., children who had
previously lived with the other parent moved to live full-time in the
recipient household. Table 5 shows the results of censoring those
households in the year when they ended up living in overcrowded
conditions. Since the models incorporating frailty suggest that unobserved
heterogeneity is not significant in this case and therefore produces results
almost identical to those of the model not incorporating frailty, we present
only the results of the complementary log-log model.
Table 5. Estimation results from discrete-time proportional hazards
model (exponentiated coefficients), for the risk of moving into
overcrowding housing (other types of adjustments into
overcrowded housing were censored at the time of the
adjustment) (no. of households 58,168, failures 7350).
Exp. coeff.
(Std errors)
Reform variable
T*[d97] (t)
1.041 ***
(0.011)
Decrease in housing allowance
0.974 ***
(0.006)
Year dummy, 1997–1999 [d97] (t)
0.908
(0.037)
Household characteristics
Age (t)
0.969 ***
(0.002)
Two children (t)
2.077 ***
(0.053)
More than two children (t)
4.520 ***
(0.200)
Housing expenses/income, 1996
1.313 ***
(0.031)
Housing market characteristics
Tobin’s
q
(t)
0.443 ***
(0.021)
Duration dependence
Time (t)
1.705 ***
(0.104)
Time^2 (t)
0.927 ***
(0.008)
Log likelihood
–29,920.1
Notes
: *** Significant at the 1% level. (t) indicates time-varying variables. Clustered
sandwich estimator of standard errors within parentheses.
